--- Begin Message ---
Hello list.

I hope you have a great day. Please find below a brief overview of some
current continuous integration matters.

The Linux/s390x problem space is now covered with the IBM Z CI service
provided by Oregon State University Open Source Lab. It uses Jenkins
and is not in the same loop as most other builds (i.e. no automatic
github.com commit status updates, at least for now), but people will
still get e-mails when things break.

The Buildbot configuration at ci.tcpdump.org has been growing, it
currently includes some permutations of Linux, BSD, 32-bit/64-bit ARM
and AMD64. Harald Welte of sysmocom has sponsored and hosted three
RPIs, which currently run about half the builds of this infrastructure.

The most useful view of the existing Buildbot configuration is probably
this (it has tags and filtering): https://ci.tcpdump.org/#/builders

Since tcpdump and libpcap originate in the Unix-like systems world, it
is remarkable that this is the first time CI covers Linux, FreeBSD,
NetBSD and OpenBSD on a given architecture: in this sense AArch64 is
fully covered by means of Buildbot and AMD64 is split between Buildbot
and Cirrus CI.

That said, adding NetBSD and OpenBSD had chased a few bugs out, some of
which I managed to fix quickly, but a few remain. To get to a starting
point with all builds passing, it had taken to omit some matrix
elements.

In particular, NetBSD has two libpcap libraries (one that comes with
the system and another that comes from pkgsrc), slightly different from
each other and from the upstream libpcap. Linking with these libraries
turned out to be difficult, so for the time being tcpdump on NetBSD
builds with local libpcap only and tcpdump build matrix is 2 times
smaller than usual.

On OpenBSD there is a slightly different issue, in that tcpdump always 
links with the system libpcap.so, but the latter is too old to pass the
tests. Also tcpdump fails to link with ../libpcap/libpcap.so when
tcpdump is compiled using GCC (libpcap is always compiled using the
default compiler, which in this case is Clang). Also CMake fails to
build tcpdump properly. This way, tcpdump build matrix is 4 times
smaller. Also tcpslice produces too many warnings and fails the build.

I cannot remember exact details now, but they are recorded in the
failed build logs, and as soon as the build matrix is the usual size
again, the build issues should manifest. It would be great if someone
could look into these issues, identify the reasons for failure and tell
what can be fixed easily.

-- 
    Denis Ovsienko

--- End Message ---
_______________________________________________
tcpdump-workers mailing list
tcpdump-workers@lists.tcpdump.org
https://lists.sandelman.ca/mailman/listinfo/tcpdump-workers

Reply via email to